A calculator from the Society of Actuaries

[nfs]

SOA calculator

 

[Nords]

Well, ooookay...

... but how does it compare to FIREcalc or Financial Engines?

(I'll check back after morning surf...)

 

[cute fuzzy bunny]

You'll need "comdlg32.ocx" to run this program.

A common problem when developers distribute programs without their needed run-time files.

For windows XP, download this file to c:\windows\system32
http://www.ascentive.com/support/new/images/lib/COMDLG32.OCX

And then click 'start', then 'run', then paste this into the run box
regsvr32 \windows\system32\COMDLG32.OCX

If you have another version of windows, follow these directions...
http://www.ascentive.com/support/new/support_dll.phtml?dllname=COMDLG32.OCX

It does let you set up EVERYTHING before giving you the run time error when you hit calculate.

Minor bitch: yet another retirement calculator that doesnt let you specify an age under 50.

 

[sgeeeee]

Hi nfs,

Thanks for the link. I didn't have any problem downloading and running the program. It has some nifty features and builds in methods to calculate your life expectancy. If you are interested in adding an annuity to your investments, the calculator provides a detailed analysis that will help you pick out the right annuity. It would appear to have all the weaknesses and limitations of any other monte carlo simulator, but I intend to play around with it and see if I find a good use for some of it's unique features.

 

[Pennhaven]

It would appear to have all the weaknesses and limitations of any other monte carlo simulator, but I intend to play around with it and see if I find a good use for some of it's unique features.

There is a quite favorable review of this calculator in today's Wall Street Journal (August 31, 2004 Page D2).

Another thing that distinguishes the Retirement Probability Analyzer is the avoidance of Monte Carlo Simulations... Mr. Milevsky's calculator relies on partial differential equations, which he says are sophisticated enough to remove the need for Monte Carlo Simulations.

SOA Retirement Probability Analyzer

 

[sgeeeee]

Another thing that distinguishes the Retirement Probability Analyzer is the avoidance of Monte Carlo Simulations... Mr. Milevsky's calculator relies on partial differential equations, which he says are sophisticated enough to remove the need for Monte Carlo Simulations.

Yes, but the inputs to the calculator are the same as monte carlo (ie. assumed means and standard deviations for rates of return without correlation between them). With that input, you have to end up with results that ignore correlations (like monte carlo).

On the other hand, like monte carlo, the user can specify whatever mean return rates and variations that they wish to investigate how they would fair under their own assumed future. That capability is difficult to achieve with historical simulators.

 

[Pennhaven]

Not trying to be critical, Guru, I just felt a factual correction may be in order, as both the Journal article and the FAQ included with the program imply that there may be concern about the validity of using Monte Carlo Scenarios for this purpose. If such concerns are valid, then stating that this model utilizes Monte Carlo Scenarios may be falsely prejudicial.

My own studies of mathematics are years in the past, so I don't have an opinion either way.

 

[sgeeeee]

Not trying to be critical, Guru, I just felt a factual correction may be in order, as both the Journal article and the FAQ included with the program imply that there may be concern about the validity of using Monte Carlo Scenarios for this purpose. If such concerns are valid, then stating that this model utilizes Monte Carlo Scenarios may be falsely prejudicial.

My own studies of mathematics are years in the past, so I don't have an opinion either way.  

I didn't mean to claim that this was a monte carlo simulation, only that it suffers from the same limitations in terms of interpretation of the final solution.

I only skimmed the manual and played around with the simulator for several minutes. But it looks to me like he starts with assumed distributions and writes PDEs to describe their evolution (magnitude drift and diffusion) through time. So instead of simulating random number generators and selection criteria for each of 30 years (or however long you assume), he applies his boundry condtions and solves the PDEs. He ends up with final probability distributions. This is a slick and elegant way to approach the problem and avoids the long simulation time issues that can plague monte carlo. But, at best, it will provide a more efficient way to find the solution that monte carlo gives you with a very large number of trials.

I could be wrong about the details because there isn't much information in the documents you can download. But 1) the simulator does not make use of historical data, 2) it does make use of assumed distributions of returns, 3) it does not have any inputs to account for correlation between the input distributions. The references might offer greater insight if someone wants to dig into the guts of the program. I might get around to that some day, but not today.

 

[sgeeeee]

Here's a link with some of the math behind this simulator. Unless you are a very serious student of partial differential equations, you won't want to even look at this stuff. The simulator does generally what I expected, but I had forgotten how complex this kind of development could be. I've decided that I am retired now and trying to dig through these details looks way too much like work.

http://www.ifid.ca/pdf_workingpapers/WP2003OCT15.pdf